Skip to content

Prompts Management

Prompts are the building blocks of your AI interactions. This section allows administrators to manage the lifecycle of these prompts, from creation to performance analysis.

The main view displays all available prompts for the current tenant.

  • Search: Find specific prompts by their ID or content.
  • Actions: Create new prompts, edit existing ones, or view their specific statistics.
  • API Reference: GET /api/v1/admin/prompts

2. Prompt Editor (CRUD)

The editor provides granular control over prompt behavior.

Content & Configuration

  • Content Editor: Edit the Thymeleaf template used to generate the prompt dynamically. See The Templating Engine for syntax details.
  • Model Configuration: Set the default LLM provider (e.g., openai) and model (e.g., gpt-5.1). See Choosing the Right Model for guidance.
  • Capabilities: Toggle flags based on the prompt's needs:
    • Reasoning: Enable or disable reasoning capabilities.
    • Multi-modal: Flag if the prompt requires image inputs.
    • Function Calling: Flag if the prompt triggers external tools.

ROI Settings

  • Time Saved: Define an estimated "Time Saved per usage" (in seconds). This value is used to calculate the Return on Investment (ROI) metrics across the platform.

3. Prompt Statistics

Each prompt has a dedicated statistics view with performance metrics:

  • Usage Count: The total number of times this prompt has been triggered.
  • Feedback: A breakdown of user ratings (Good, Bad, Neutral) to assess output quality.
  • Cost & ROI: Token consumption (totalCost, costAverage) and the specific time saved by this prompt.

4. Prompt Tester

The Prompt Tester allows you to execute a prompt directly from the admin interface — without needing to set up a conversation or use external tools like cURL.

How It Works

  1. Variable Detection: When you open the tester for a prompt, it automatically parses the Thymeleaf template and detects all variables (e.g., ${documentId}, ${language}).

  2. Variable Configuration: For each detected variable, an input field is displayed. You can provide:

    • Text values — For string variables like document IDs, language codes, or user queries.
    • Image values — For multi-modal prompts that expect Base64-encoded images.

  3. Execute: Click "Test" to send the prompt with the configured variables to the LLM. The system uses the prompt's defaultLlmProvider and defaultLlmModel settings.

  4. Results: The tester displays:

    • The LLM response.
    • Token usage (input/output).
    • Response latency.

  5. cURL Generation: The tester generates the equivalent cURL command for the test configuration, making it easy to reproduce the request from a terminal or integrate into scripts.

Iterate Quickly

Use the Prompt Tester to refine your prompt content and model selection before deploying to production. Adjust the template, change variables, and re-test — all without leaving the admin panel.